15.29. __nv_fp8x4_ue5m3

struct __nv_fp8x4_ue5m3

This structure implements the datatype for storage and operations on the vector of four fp8 values of ue5m3 kind each: with no sign, 5 exponent, 1 implicit and 3 explicit mantissa bits. The encoding doesn’t support Infinity. NaNs are limited to 0xFF values.

Public Functions

__nv_fp8x4_ue5m3() = default: Constructor by default.

__host__ __device__ inline explicit __nv_fp8x4_ue5m3(const __half2 flo, const __half2 fhi)

Constructor from a pair of __half2 data type values, relies on __NV_SATFINITE behavior for large input values and cudaRoundNearest for rounding.

See also

__nv_cvt_float_to_ue5m3 for further details

__host__ __device__ inline explicit __nv_fp8x4_ue5m3(const __nv_bfloat162 flo, const __nv_bfloat162 fhi)

Constructor from a pair of __nv_bfloat162 data type values, relies on __NV_SATFINITE behavior for large input values and cudaRoundNearest for rounding.

See also

__nv_cvt_bfloat16raw_to_ue5m3 for further details

inline explicit __NV_SILENCE_DEPRECATION_BEGIN __host__ __device__ __nv_fp8x4_ue5m3(const double4 f)

Constructor from double4 vector data type, relies on __NV_SATFINITE behavior for large input values and cudaRoundNearest for rounding.

See also

__nv_cvt_double_to_ue5m3 for further details

inline explicit __NV_SILENCE_DEPRECATION_END __host__ __device__ __nv_fp8x4_ue5m3(const double4_16a f)

Constructor from double4_16a vector data type, relies on __NV_SATFINITE behavior for large input values and cudaRoundNearest for rounding.

See also

__nv_cvt_double_to_ue5m3 for further details

__host__ __device__ inline explicit __nv_fp8x4_ue5m3(const double4_32a f)

Constructor from double4_32a vector data type, relies on __NV_SATFINITE behavior for large input values and cudaRoundNearest for rounding.

See also

__nv_cvt_double_to_ue5m3 for further details

__host__ __device__ inline explicit __nv_fp8x4_ue5m3(const float4 f)

Constructor from float4 vector data type, relies on __NV_SATFINITE behavior for large input values and cudaRoundNearest for rounding.

See also

__nv_cvt_float_to_ue5m3 for further details

__host__ __device__ inline explicit operator float4() const: Conversion operator to float4 vector data type.

Public Members

__nv_fp8x4_storage_t __x: Storage variable contains the vector of four fp8 floating-point data values.